Stochastic Discount Factor

Wang Wei Mun

Lee Kong Chian School of Business

Singapore Management University

August 29, 2022

Wang Wei Mun Stochastic Discount Factor 1 / 23

Consumption Choice Model

Investor receives utility from regular consumption of goods

and services, which is financed by investor’s existing wealth
Consider static (or “one-period”) model in which investor only
consumes at start and end of single time period
Investor starts with initial wealth of W0 and immediately
consumes C0 , which leaves remaining wealth of (W0 − C0 )
Investor can invest remaining wealth in any of n risky assets:
i’th asset has initial price of Pi and (random) final value of X̃i
=⇒ (random) return of R̃i = X̃i /Pi
One of the risky assets may in fact be riskless bond with
(non-random) risk-free rate of Rf

Wang Wei Mun Stochastic Discount Factor 2 / 23

Portfolio Choice & Budget Constraint

Suppose that investor invests proportion wi ofPremaining

n
wealth into i’th asset, subject to constraint: i=1 wi = 1
Investor’s final wealth depends on realised portfolio return:
n
X
W̃1 = (W0 − C0 ) wi R̃i
i=1

No further opportunity for consumption after end of time

period, so investor optimally chooses to consume final wealth:
n
X
C̃1 = (W0 − C0 ) wi R̃i
i=1

Wang Wei Mun Stochastic Discount Factor 3 / 23

Investor’s Utility

Investor’s overall utility of consumption

 will depend on both
initial and final consumption: V C0 , C̃1
For simplicity, assume that investor has time-separable utility
of consumption =⇒ investor’s utility of consumption at given
point of time is not affected by past or future consumption:
  h  i
V C0 , C̃1 = U(C0 ) + δE U C̃1

Here δ ∈ (0, 1) is subjective discount factor that reflects

investor’s rate of time preference (i.e., impatience)
Assume that U(·) is strictly increasing and concave =⇒
investor will be non-satiated and risk averse

Wang Wei Mun Stochastic Discount Factor 4 / 23

Consumption and Portfolio Choice Problem

At start of time period, investor chooses initial consumption

of C0 and portfolio weights of wi (for investment portfolio) so
as to maximise overall utility, subject to relevant constraints:

n
( !)
h  i X
max L = U(C0 ) + δE U C̃1 + λ 1 − wi
C0 ,{wi }
i=1

First-order optimality condition for initial consumption, after

applying chain rule since C̃1 is function of C0 :

n
" #
∂L ′
 X
=0 =⇒ U (C0∗ ) = δE U ′
C̃1∗ wi∗ R̃i
∂C0
i=1

Wang Wei Mun Stochastic Discount Factor 5 / 23

Optimal Asset Allocation – Part 1

First-order optimality conditions for portfolio weights, after

applying chain rule since C̃1 is function of wi :

∂L
= 0 =⇒
∂wi
h   i λ
δE U ′ C̃1∗ R̃i = ∀ i = 1, . . . , n
W0 − C0∗

All assets must have same expected marginal-utility-weighted

return, based on marginal utility of optimal final consumption:
h   i h   i
E U ′ C̃1∗ R̃i = E U ′ C̃1∗ R̃j ∀ i, j

Wang Wei Mun Stochastic Discount Factor 6 / 23

Optimal Asset Allocation – Part 2

Additional dollar invested in i’th assetproduces

 return of R̃i ,
′
which provides additional utility of U C̃1 R̃i when consumed
If i’th asset provides higher expected marginal-utility-weighted
return, then investor will shift investment into i’th asset
More investment in i’th asset leads to higher correlation
 
between R̃i and C̃1 =⇒ C̃1 tends to higher and U ′ C̃1
tends to be lower when R̃i is above average, and vice versa
Hence expected marginal-utility-weighted return for i’th asset
will fall since utility function is concave =⇒ U ′′ (·) < 0
Investor will shift investment across risky assets until all assets
have same expected marginal-utility-weighted return

Wang Wei Mun Stochastic Discount Factor 7 / 23

Intertemporal Allocation

Use equality of expected marginal-utility-weighted returns to

simplify optimality condition for initial consumption:
n
X  h   i h   i
′
U (C0∗ ) = wi∗ δE U ′ C̃1∗ R̃i = δE U ′ C̃1∗ R̃i
i=1

LHS represents marginal utility from one unit of initial

consumption, while RHS represents discounted expected
marginal utility from R̃i units of final consumption
Hence investor will shift between initial consumption and
investment in final consumption to equalise marginal benefit
Applies to all assets, as well as any combination of assets

Wang Wei Mun Stochastic Discount Factor 8 / 23

Asset Pricing Formula

Rearrange to get asset pricing formula:

       
U ′ C̃1∗ U ′ C̃1∗
E δ R̃i  = 1 =⇒ Pi = E δ X̃i 
U ′ (C0∗ ) U ′ (C0∗ )
 
Here M̃ = δU ′ C̃1∗ /U ′ (C0∗ ) > 0 represents investor’s
intertemporal marginal rate of substitution (IMRS)
Hence investor’s IMRS acts as pricing kernel (or stochastic
discount factor) that relates initial price to final value:
h i h i
E M̃ R̃i = 1 =⇒ Pi = E M̃ X̃i

Wang Wei Mun Stochastic Discount Factor 9 / 23

Consumption CAPM – Part 1

Assume that riskless bond exists:

h i h i
E M̃Rf = 1 =⇒ E M̃ = Rf−1 > 0

Expand expectation of product in asset pricing formula:

h i h i h i h i
E M̃ R̃i = E M̃ E R̃i + Cov M̃, R̃i = 1

Rearrange to get pricing formula for Consumption CAPM:

h i h   i
h i Cov M̃, R̃i Cov U ′ C̃1∗ , R̃i
E R̃i − Rf = − h i =− h  i
E M̃ E U ′ C̃1∗

Wang Wei Mun Stochastic Discount Factor 10 / 23

Consumption CAPM – Part 2

 
Suppose that R̃i has negative correlation with U ′ C̃1∗
Implies that asset return tends to be high when marginal
utility of final consumption is low, and vice versa
Hence investor is likely to receive more consumption when
consumption is less valuable, and vice versa
Asset has undesirable payoff characteristics, so investor will
demand large risk premium for holding this “risky” asset
 
Conversely, if R̃i has positive correlation with U ′ C̃1∗ , then
investing in asset provides insurance against low consumption,
so investor is willing to accept negative risk premium

Wang Wei Mun Stochastic Discount Factor 11 / 23

Consumption CAPM → CAPM

Suppose that investor’s optimal portfolio is affine combination

of market portfolio and riskless asset
Market return will have perfect negative correlation with
investor’s marginal utility of final consumption, and pricing
kernel will be linear function of market return
Market risk becomes only source of systematic risk, so
Consumption CAPM will give same pricing formula as CAPM
Requires that investor has quadratic utility of consumption, or
that all risky assets have normal returns
Hence investor will become satiated, and then investor’s
marginal utility will drop below zero, as consumption rises

Wang Wei Mun Stochastic Discount Factor 12 / 23

Volatility Bound – Part 1
h i h i
Let µM = E M̃ = Rf−1 and Cov M̃, R̃i = ρσM σi , where ρ is
correlation coefficient between M̃ and R̃i
Apply to pricing formula for Consumption CAPM:
h i
h i ρσM σi E R̃i − Rf σM
E R̃i − Rf = − =⇒ = −ρ
µM σi µM

Use ρ ∈ [−1, 1] to get Hansen–Jagannathan (H–J) bound:

h i
E R̃i − Rf σM
≤
σi µM

Wang Wei Mun Stochastic Discount Factor 13 / 23

Volatility Bound – Part 2

LHS of H–J bound is Sharpe ratio of any risky asset, while

RHS of H–J bound is “volatility ratio” for pricing kernel
But H–J bound also applies to any portfolio, since pricing
formula for Consumption CAPM applies to any portfolio
Hence volatility ratio of pricing kernel cannot be less than
highest Sharpe ratio out of all possible portfolios
Annual risk premium of around 7% and annual standard
deviation of around 17% for U.S. stock market returns =⇒
Sharpe ratio of around 0.4, so pricing kernel is highly volatile
Pricing kernel has lower limit of zero but no upper limit =⇒
probability distribution should be heavily skewed on right side

Wang Wei Mun Stochastic Discount Factor 14 / 23

Power Utility – Part 1

Consider investor with power utility of consumption:

C 1−γ
U(C ) = =⇒
1−γ
!−γ " !#
C̃1∗ C̃1∗
M̃ = δ = δ exp −γ ln
C0∗ C0∗

Suppose that optimal consumption growth has lognormal

distribution with mean of µc and variance of σc2 :
!
C̃1∗
ln = µc + σc z̃, z̃ ∼ N(0, 1)
C0∗

Wang Wei Mun Stochastic Discount Factor 15 / 23

Power Utility – Part 2

Apply result for variance of pricing kernel to volatility ratio:

h i h i h i2
Var M̃ = E M̃ 2 − E M̃
2
=⇒ σM = µM 2 − µ2M
 1
σM µM 2 2
=⇒ = 2
−1
µM µM

Apply results for lognormal random variable:

h i 1 2 2
µM = δE e −γ(µc +σc z̃) = δe −γµc + 2 γ σc = η
h i 2 2 2 2
µM 2 = δE e −2γ(µc +σc z̃) = δe −2γµc +2γ σc = η 2 e γ σc

Wang Wei Mun Stochastic Discount Factor 16 / 23

Power Utility – Part 3

Substitute for µM and µM 2 in equation for volatility ratio of

pricing kernel, and apply e x ≈ 1 + x for small values of x:
 1  1
σM µM 2 2
γ 2 σc2 2
= 2
− 1 = e − 1 ≈ γσc
µM µM

Now substitute into result for H–J bound:

h i
σM E R̃i − Rf
≈ γσc ≥
µM σi

Wang Wei Mun Stochastic Discount Factor 17 / 23

Equity Premium Puzzle

Sharpe ratio of around 0.4 for U.S. stock market

σc ≈ 2% based on real annual per capita consumption for
post-war U.S. economy (i.e., after World War II)
Hence investor with power utility who consumes real per
capita consumption must have γ ≳ 20, which is generally
considered as unreasonably high degree of relative risk aversion
Equity premium puzzle: investor with time-separable power
utility of consumption and lognormal consumption growth
must have unreasonably high degree of relative risk aversion
Either investors don’t have power utility of consumption, or
consumption growth doesn’t have lognormal distribution

Wang Wei Mun Stochastic Discount Factor 18 / 23

Skewness Bound

Investor’s optimal consumption growth has lognormal

distribution =⇒ small amount of negative (left) skewness
So for investor with power utility of consumption, distribution
for pricing kernel will have positive (right) skewness that
increases with investor’s (relative) risk aversion
Empirical evidence suggests that probability distribution for
pricing kernel should have large amount of positive skewness
Hence investor must also have high degree of relative risk
aversion to satisfy “skewness bound” for pricing kernel
But what if empirical data on post-war consumption
understates volatility and skewness of consumption growth?

Wang Wei Mun Stochastic Discount Factor 19 / 23

Rare Disasters – Part 1

Now suppose that optimal consumption growth also contains

random variable that represents effect of rare disasters:
!
C̃1∗
ln = µc + σc z̃ + ν̃,
C0∗

ln ϕ with probability of π
ν̃ =
0 with probability of 1 − π

Here π ∈ [0, 1] is probability that rare disaster occurs

Then 1 − ϕ is fraction of optimal consumption that is lost in
event of disaster, where ϕ ∈ (0, 1)
For simplicity, assume that ν̃ is independent of z̃

Wang Wei Mun Stochastic Discount Factor 20 / 23

Rare Disasters – Part 2

Disasters are events that result in great economic disruption,

such as Great Depression, World War I, and World War II
Other examples are outbreak of infectious and deadly viral
pandemic, or asteroid striking Earth in densely populated area
Historical data usually covers time periods without disasters,
which makes consumption growth appear less volatile
Moreover, excluding disasters severely understates amount of
negative (left) skewness in consumption growth
Robert Barro conducted survey of major disasters of 20th
century, and concluded that π = 1.7% and ϕ = 0.65 for real
annual per capita consumption growth

Wang Wei Mun Stochastic Discount Factor 21 / 23

Rare Disasters – Part 3

Apply results for lognormal random variable:

µM = ηE e −γ ν̃ = η 1 + π ϕ−γ − 1
  

2 2  2 2 
µM 2 = η 2 e γ σc E e −2γ ν̃ = η 2 e γ σc 1 + π ϕ−2γ − 1


Can also apply 1 + x ≈ e x as long as γ is reasonably small:

−γ −1
µM ≈ ηe π(ϕ )
2 2 −2γ
µM 2 ≈ η 2 e γ σc +π(ϕ −1)
 1   12
σM µM 2 2
γ 2 σc2 +π (ϕ−γ −1)
2
=⇒ = − 1 ≈ e − 1
µM µ2M

Wang Wei Mun Stochastic Discount Factor 22 / 23

Rare Disasters – Part 4

If γ is reasonably small, then γ 2 σc2 ≈ 0, so apply e x ≈ 1 + x

to remaining terms:

2 √ 1
σM  −γ 2
≈ e π(ϕ −1) − 1 ≈ π ϕ−γ − 1


µM

No equity premium puzzle since γ ≳ 3.3 (based on Sharpe

ratio of U.S. stock market), which represents acceptable
degree of relative risk aversion:
σM √
≈ 0.017 0.65−3.3 − 1 = 0.41


γ = 3.3 =⇒
µM

Wang Wei Mun Stochastic Discount Factor 23 / 23